1. Associative Memory: A Spiking Neural Network Robotic Implementation
André Cyr, Frédéric Thériault, Matt Ross, Sylvain Chartier

2. Associative Memory (AM)
Autoassociative (pattern completion):
Input
Output
Input
Output
Heteroassociative:
Input
Output

3. Associative Memory (AM) Cont.
AM usually refers to complex brain structures [1,2].
Generally modeled at the phenomenological level.
Why model at all?
Cognitive economy for natural or artificial agents [3].
Artificial Neural Network approach
SOM maps, especially for robotic navigation [4,5]

4. Spiking Neural Network (SNN) Model
Addition of a temporal dimension for encoding information.
Has become another approach to modeling the AM phenomenon [6,7,8].
Current objective: Create a simple embodied bio- inspired mechanism for the AM model, by exploiting the inherent computational and temporal features of a SNN model.
Analytical description of neuronal spike model dynamics can be found at:

5. Spiking Neural Model for AM
Temporal neural features
Asymmetric spike timing
Asymmetric spike-timing dependent plasticity (STDP) learning function [9]:

6. Architecture

7. Learning Mechanism
graphics A-B-C image caption of the three black dots on the left column generates single spikes of their respective neurons.
graphics D-E-F The randomized spike delays (PSP) between the input and the associative neural layer are produced (graphics G-H-I).
These small PSP delays are enough to be used from the STDP rule to adjust the modulating synaptic factor (graphics J to O: percentage scale). Since the synaptic delays are randomized, anti-hebb happens half of the time, but because there was a positive bias favoring the hebb side, eventually results in an increased weight and affects the spiking response of the associative neural units.
graphics P-Q-R output spikes to the LCD device.

8. Autoassociative Simulation

9. Autoassociative Simulation Cont.

10. Heteroassociative Simulation

11. Robotic Implementation
Full video available at:

12. Conclusion
With simple visual tasks and minimalist cellular circuits, it was shown that asymmetric synaptic delays and asymmetric STDP learning are sufficient conditions to achieve pattern- completion and noise tolerance for auto and heteroassociative tasks.

13. References
Rolls, E.: The mechanisms for pattern completion and pattern separation in the hippocampus. Front. Syst. Neurosci 7(74), (2013)
Smith, D., Wessnitzer, J., Webb, B.: A model of associative learning in the mushroom body. Biological cybernetics 99(2), (2008)
Chartier, S., Giguere, G., Langlois, D.: A new bidirectional heteroassociative memory encompassing correlational, competitive and topological properties. Neural Networks 22(5), (2009)
Touzet, C.: Modeling and simulation of elementary robot behaviors using associative memories. International Journal of Advanced Robotic Systems 3(2), (2006)
Tangruamsub, S., Kawewong, A., Tsuboyama, M., Hasegawa, O.: Self- organizingincremental associative memory-based robot navigation. IEIC TRANSACTIONS on Information and Systems 95(10), (2012)
Gerstner, W., van Hemmen, J.: Associative memory in a network of ‘spiking’ neurons, Network. Computation in Neural Systems, 3:2, (1992)
Tan, C., Tang, H., Cheu, E., Hu, J.: A computationally ecient associative memory model of hippocampus ca3 by spiking neurons. In: Neural Networks (IJCNN), The International Joint Conference on. pp. 1{8. IEEE (2013)
Jimenez-Romero, C., Sousa-Rodrigues, D., Johnson, J.: Designing behaviour in bio-inspired robots using associative topologies of spiking-neural-networks. arXiv preprint arXiv: (2015)
Dan, Y., & Poo, M. (2004). Spike Timing-Dependent Plasticity of Neural Circuits. Neuron, 44(1), 23–30.